I am reading Gilbarg/Trudinger Chapter 9. Now on Theorem 9.11 page 235 reprint of the 1998 edition. And as usual I am stuck in on of the assertions of the proof.

It says that given $\Omega\subset \mathbb{R}^n$ bounded domain, and for every $v\in W_0^{2,p}(\Omega)$, $1<p<\infty$, on can obtain the estimate \begin{equation*} ||D^2v||_{p;\Omega}\leq \frac{C}{\lambda}||L v||_{p,\Omega} \quad \text{with } C=C(n,p) \end{equation*}

Where $Lv := a^{ij}D_{ij}v$ (Einstein summation convention) is a uniformly elliptic differential operator with constant coefficients and $\lambda$ is the constant such that $a^{ij}\xi_i\xi_j\geq\lambda|\xi|^2$.

In order to prove this the book reccomends to use two previous results,

1) Under the same hypothesis on $v$ be have the estimate (Corollary 9.10)
\begin{equation*} ||D^2v||_{p;\Omega}\leq C||\Delta v||_{p,\Omega} \end{equation*}

2) One can use a linear transformation $Q$ (This idea comes from Lemma 6.1 in the book) to define a new function $\tilde{v}$ such that \begin{equation*} a^{ij}D^{ij}v(x) = \tilde{a}^{ij}\tilde{v}(y) \end{equation*} where $y=xQ$.

I tried to use this two results by taking $\tilde{v}$ such that $\tilde{a}^{ij}\tilde{v}(y)=\Delta \tilde{v}$ and then seeing how are the relations between the norms of $v$ and $\tilde{v}$ involved in the estimate but could not get to any successful calculation. I'll add those calculations if necessary. But if you have a hint on a different, successful, approach it is welcomed.

  • $\begingroup$ This is the Calderon zygmund inequality my friend it is not a easy joke:) $\endgroup$ – Guy Fsone Nov 10 '17 at 19:22
  • $\begingroup$ What is $L_0$? specify that $\endgroup$ – Guy Fsone Nov 10 '17 at 19:25
  • $\begingroup$ @GuyFsone I changed the notation to simply $L$ because the subindex was unnecesary. It is a constant coefficients operator. It is used to prove the general non coefficient estimate. The part I don't get is this one exactly, that apparently is not so difficult since is just the constant coefficients case. It is assumed that we have the $L^p$ bounds for the Newtonian potential (That is the same bound that we want but in the special case in which our $L$ is just the Laplacian). This assumed result is what I believe you called the Calderon-Zygmund inequality and Corollary 9.10 in the book. $\endgroup$ – I.C. Nov 11 '17 at 6:22
  • $\begingroup$ I edited the question so the previous results are easily noted. They can be assumed and used. I think I do understand the demonstrations of those results. $\endgroup$ – I.C. Nov 11 '17 at 6:24

Lemma 6.1 describes in detail how to find a linear change of coordinates $y=xQ$ so that $\bar v(y) = v(x(y))$ satisfies $$(\Delta \bar v)(y) = (Lv)(x(y)),$$ so that the estimate for the Laplacian gives $$\|D^2 \bar v \|_p \le C(n,p) \| \Delta \bar v \|_p.\tag{1}$$

We can now rewrite both sides in terms of $v$ to obtain the desired estimate. The RHS is easy: taking the $p^\text{th}$ power for convenience and applying the change of variables formula for integration we find

$$ \| \Delta \bar v \|_p^p = \int |\Delta \bar v(y)|^p dy=\int |Lv(x)|^p\left| \det \frac{dy}{dx}\right|dx = \left|\det Q\right| \|L v \|^p_p.\tag{2}$$

The LHS is a little trickier, since in addition to the volume transformation we need to turn the $y$ derivatives into $x$ derivatives. From $y = xQ$ we find $D_{x_i} = Q_{ij}D_{y_j},$ so $|D_x^2 v| \le C(n)|Q|^2|D_y^2 \bar v|.$ Recalling (from 6.1) that $|Q| \le C(n) \lambda^{-1/2}$ and doing the change of variables we find $$\| D^2 v \|_p^p = \int |D_x^2 v|^p dx \le \left|\det Q\right|^{-1}\lambda^{-p}\|D^2 \bar v\|_p^p;$$ so combining with $(1)$ and $(2)$ we find $$\|D^2 v\|^p_p \le \lambda^{-p}C(n,p)\|Lv\|_p^p$$ as desired. (I've been vague about which matrix norm I'm using here, but it doesn't matter - they're all equivalent with constant depending only on $n$.)


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.